import json from typing import Dict, Any, List from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import JsonOutputParser from pydantic import BaseModel, Field from backend.core.errors import AppError from backend.core.logging import get_logger from backend.integrations.llm.base import BaseLLMClient logger = get_logger(__name__) # Using Pydantic to structure LangChain output for step execution class APIRequest(BaseModel): method: str url: str headers: Dict[str, Any] = Field(default_factory=dict) query_params: Dict[str, Any] = Field(default_factory=dict) form_data: Dict[str, Any] = Field(default_factory=dict) body: Any = Field(default_factory=dict) files: Dict[str, str] = Field(default_factory=dict) # NEW: {fieldName: fileName} class StepAnalysis(BaseModel): required_inputs: List[Any] = Field(default_factory=list) values_from_previous_steps: List[Any] = Field(default_factory=list) generated_test_data: List[Any] = Field(default_factory=list) class StepExecutionPlan(BaseModel): step: int request: APIRequest analysis: StepAnalysis # Prompt template for step preparation STEP_PROMPT = """ You are an API execution assistant. Your job is to build a precise HTTP request for the CURRENT step. NOTATION IN SPEC: - {{prop*(str)}}: 'prop' is a REQUIRED string. - {{prop(int)}}: 'prop' is an OPTIONAL integer. - {{status(str[active|inactive])}}: 'status' is a string that MUST be one of the enum values [active, inactive]. - {{date(string:date-time)}}: 'date' is a string in date-time format. INPUTS: 1. API Spec (current endpoint): {endpoint_spec} 2. Base URL: {base_url} 3. Scenario: {scenario} 4. Current Step: {step_definition} 5. User Credentials: {user_credentials} 6. Context Variables (Priority): {context_variables} 7. Assets (Semantic Context): {available_assets} CORE EXECUTION RULES: 1. STRICT SCHEMA ADHERENCE (CRITICAL): - You MUST ONLY use parameters, headers, and body fields that are explicitly defined in the API Spec. DO NOT hallucinate variables or fields that do not exist. - Respect data types. If an integer is expected, output an integer, not a string. 2. URL & PATH PARAMETERS: - You MUST replace all path variables like {{id}} or {{bus_id}} with actual values from Context or Step Definition. - Example: If URL is "/buses/{{bus_id}}" and Context has bus_id=101, result URL must be "/buses/101". 3. AUTHORIZATION: - Use credentials from "User Credentials" or "Context Variables" to satisfy authentication (token, api_key, etc). - Only include an "Authorization" header if required by the spec. Use the correct format (e.g., "Bearer ", "Token ", or "Basic ") as needed. - DO NOT default to "Bearer" if the credential type is not specified. 4. DATA SOURCE PRIORITY: - Priority 1: Context Variables (from previous steps). - Priority 2: User Credentials (for login/auth). - Priority 3: Step Definition / Scenario data. 5. STRICT ENUM ADHERENCE: - If a field has an enum like [A|B|C], you MUST pick one of those exact values. 6. FORM vs JSON: - Content-Type: application/x-www-form-urlencoded -> use 'form_data'. - Content-Type: application/json -> use 'body'. - If the endpoint receives a JSON array, the 'body' field must be an array instead of a dict. 7. FILES: Use 'files' object {{"field": "filename"}} using names from the provided Assets. 8. NO TEMPLATE STRINGS: - Do not leave {{var}} or syntax in the final 'request' JSON. Fully resolve variables. 9. COOKIE RULE: - Do NOT add a Cookie header unless the spec explicitly requires it. Bearer token auth does not use cookies. 10. PHP SINGLE-FILE API RULE: - When the Base URL path ends with a .php filename (e.g. mobile_api.php), the `action` routing parameter MUST be in `query_params` (e.g. {{"action": "login", "v": 1}}). - For POST/PUT/PATCH requests you may also include `action` in the JSON body, but it is NOT needed for GET requests. - NEVER append the semantic OpenAPI path (e.g. /auth/login) after the .php filename — the URL must stay exactly at the .php file and route via ?action= only. - Correct URL for login: "https://host/mobile_api.php", query_params: {{"action": "login", "v": 1}} 11. MISSING ID RULE (CRITICAL): - If a required path parameter (e.g. {{bus_id}}) or required body field is an ID that is NOT present in Context Variables and cannot be reasonably generated (it is a real DB record ID, not a UUID you create), do NOT guess a value. - Instead, set the parameter to the literal string "__MISSING____" (e.g. "__MISSING__bus_id__") so the caller can detect and block execution. - Examples of IDs you MUST NOT guess: bus_id, product_id, order_id, seat_id, route_id, ticket_id. - Examples of IDs you CAN generate: new UUIDs for user registration, random test email addresses. RETURN JSON ONLY: {{ "step": 1, "request": {{ "method": "POST", "url": "", "headers": {{"Authorization": "..."}}, "query_params": {{}}, "form_data": {{}}, "body": {{}}, "files": {{}} }}, "analysis": {{ "required_inputs": ["list_of_fields_you_filled"], "values_from_previous_steps": ["list_of_fields_taken_from_context"], "generated_test_data": ["list_of_fields_you_invented"] }} }} """ def prepare_step_request( endpoint_spec: Dict[str, Any], base_url: str, scenario: str, step_definition: Dict[str, Any], context_variables: Dict[str, Any], user_credentials: Dict[str, Any] = None, available_assets: Dict[str, Any] = None, llm: BaseLLMClient = None, ) -> StepExecutionPlan: if user_credentials is None: user_credentials = {} if available_assets is None: available_assets = {} parser = JsonOutputParser(pydantic_object=StepExecutionPlan) prompt = PromptTemplate( template=STEP_PROMPT, input_variables=["endpoint_spec", "base_url", "scenario", "step_definition", "user_credentials", "context_variables", "available_assets"], ) # ── PHASE 1: LOG EXECUTOR INPUTS ───────────────────────────────────────────── step_num = step_definition.get("step", "?") endpoint_label = step_definition.get("endpoint", "?") logger.info("="*70) logger.info("[EXECUTOR] ▶ STEP %s PREPARATION STARTED: %s", step_num, endpoint_label) logger.info("[EXECUTOR] Base URL : %s", base_url) logger.info("[EXECUTOR] Step Definition : %s", json.dumps(step_definition)) logger.info("[EXECUTOR] Credentials : %s", json.dumps(user_credentials)) logger.info("[EXECUTOR] Context Variables : %s", json.dumps(context_variables)) logger.info("[EXECUTOR] Endpoint Spec : %s", json.dumps(endpoint_spec) if isinstance(endpoint_spec, dict) else str(endpoint_spec)) logger.info("[EXECUTOR] Available Assets : %s", json.dumps(list(available_assets.keys()) if isinstance(available_assets, dict) else [])) logger.info("="*70) msg = prompt.format_prompt( endpoint_spec=json.dumps(endpoint_spec, separators=(',', ':')), base_url=base_url, scenario=scenario, step_definition=json.dumps(step_definition, separators=(',', ':')), user_credentials=json.dumps(user_credentials, separators=(',', ':')), context_variables=json.dumps(context_variables, separators=(',', ':')), available_assets=json.dumps(available_assets, separators=(',', ':')) ).to_messages() # ── PHASE 2: LLM CALL ──────────────────────────────────────────────────── logger.info("[EXECUTOR] ↑ Sending prompt to LLM...") content, token_usage = llm.invoke(msg) logger.info("[EXECUTOR] ↓ Raw LLM response:") logger.info("-"*70) logger.info("%s", content) logger.info("-"*70) # ── PHASE 3: PARSE ─────────────────────────────────────────────────────── try: parsed_response = parser.parse(content) except Exception as e: logger.warning("[EXECUTOR] ⚠ Step request parse failed, retrying: %s", e) content, token_usage = llm.invoke(msg) logger.info("[EXECUTOR] ↓ Raw LLM response (retry):") logger.info("-"*70) logger.info("%s", content) logger.info("-"*70) try: parsed_response = parser.parse(content) except Exception as e2: raise AppError(500, f"LLM returned unparseable output after 2 attempts: {e2}") # ── PHASE 4: LOG FINAL REQUEST ────────────────────────────────────────────── req = parsed_response.get("request", {}) logger.info("[EXECUTOR] ✔ Request plan generated for step %s:", step_num) logger.info("[EXECUTOR] Method : %s", req.get("method")) logger.info("[EXECUTOR] URL : %s", req.get("url")) logger.info("[EXECUTOR] Query Params: %s", json.dumps(req.get("query_params", {}))) logger.info("[EXECUTOR] Headers : %s", json.dumps(req.get("headers", {}))) logger.info("[EXECUTOR] Body : %s", json.dumps(req.get("body", {}))) logger.info("[EXECUTOR] Form Data : %s", json.dumps(req.get("form_data", {}))) logger.info("[EXECUTOR] Files : %s", json.dumps(req.get("files", {}))) logger.info("[EXECUTOR] Token usage : %s", token_usage) logger.info("="*70) return StepExecutionPlan(**parsed_response), token_usage